Research

The work behind the model — methodology, validation, and what six years of data actually support. Free, because trust is the product. RSS.

2026-07-12 · Methodology
The calibration step we kept: when making the model more accurate would have made it worse
Our research sprint found a calibration layer that hurts raw accuracy. We re-validated the pick filter without it before touching anything — and the filter died. Here's the full gate, numbers and all.
2026-07-10 · Methodology
Why our model doesn't beat Pinnacle — and why that's the point
Six pre-registered studies say we have no edge against the sharpest closing line in the world. Here's why we publish that, what calibration actually buys, and where the real product lives.

The calibration step we kept: when making the model more accurate would have made it worse